Room acoustic simulation is a computational method used to predict how sound propagates in an enclosed space. To accurately replicate the acoustic characteristics of a real-world environment, it is essential to specify appropriate boundary conditions, such as room geometry, source configurations, and surface acoustic properties. Among these, the absorption coefficient, which represents the proportion of sound energy absorbed or reflected by walls, plays a critical role in determining the acoustic response. However, accurately measuring absorption coefficients in real environments typically requires a large number of measurements. To reduce this burden, recent studies have proposed deep learning approaches for estimating absorption coefficients from microphone recordings. Most of these methods, however, estimate only the mean absorption coefficient for each wall surface. In practice, wall surfaces often exhibit spatially non-uniform absorption due to variations in materials or surface structure. In this paper, we propose a deep learning-based method for estimating frequency-dependent absorption coefficients of subdivided wall regions in rectangular rooms. The proposed model estimates the absorption coefficients for each subdivided wall region across multiple frequency bands based on time-domain sound pressure signals.
Okawa et al. (Wed,) studied this question.